MoRE improves 3D object motion from a single image
MoRE: Motion-Regularized Encoding for Articulated 3D Generation
Computer Vision and Pattern Recognition
Summary
Generating a 3D model of an object with moving parts from just one image is tough because the model must capture both how it looks and how its parts move. The authors created MoRE, a way to better represent the joints that connect parts, using a special math representation to avoid confusion about joint positions. They also trained the system to match the expected movements of these joints, helping the model better recreate how the object moves in 3D. Tests showed MoRE makes more accurate 3D shapes and joint motions than earlier methods.
What this means in practice
- •For 3d animators: Automatically generate accurately movable 3D models from single images to speed up animation workflows.
- •For augmented reality developers: Use detailed articulated 3D models from photos to create more realistic and interactive AR experiences.
Authors
Guantian Zheng, Shan Jin
Abstract
Generating an articulated object from a single image requires recovering both its appearance and the mechanism governing its parts. A compact encoding of joint attributes does not by itself ensure that reconstruction preserves the resulting motion. We present MoRE, Motion-Regularized Encoding for articulated 3D generation. MoRE represents coordinate-aligned joint lines with axis classification and two axis-orthogonal Plücker moment coefficients, removing the arbitrary choice of an origin along the axis. During representation learning, decoded joint hypotheses are used to move corresponding three-dimensional anchors, and their motions are compared with those of annotated joints. A differentiable expectation over joint types and axis directions couples attribute reconstruction to the motion it induces. An image-conditioned flow learns the resulting functional latents, while a multi-state-adapted Gaussian model supplies appearance. At inference, recovered part assignments and joints articulate a single set of rest-pose Gaussians without regenerating the object for each state. Experiments on PartNet-Mobility and ACD show stronger part-box agreement, articulated-state geometry and visual similarity than the compared baselines, while joint prediction reduces axis, origin and range errors relative to ArtiLatent. Controlled ablations support the joint-line encoding and show that motion supervision reduces anchor-motion error. Code and pretrained checkpoints will be released in the future.